<kc8apf>
LoRa gets you really impressive range. 20-30 miles easily.
<mithro>
awygle: I think someone is doing a decent job of marketing
<rqou>
<troll>but morse code over shortwave gets you the entire planet</troll>
<kc8apf>
data rate is terrible
<kc8apf>
RTTY and AX.25 can get you at least 300 baud over HF
<zkms>
oh hey have you heard of the people doing high-frequency trading over HF (like, shortwave, not microwave)
<kc8apf>
because ionosphere bounce is faster than fiber and LEO?
<zkms>
yep
<sorear>
I'm eagerly awaiting the day neutrino communication becomes possible for those with money to burn and modest data rate reqs
<zkms>
(tbf it's sorta an easier problem, they have lots of 100% reliable fibre bandwidth and GPS antennas on both sides so they can do modulation/coding-scheme negotiation, exchange channel sounding information, maintain protocol state, etc, all over the fibre link and not over the HF link)
pie__ has quit [Ping timeout: 240 seconds]
<rqou>
sorear: can't we technically already do that by turning a nuclear reactor on and off? :P
<sorear>
i suppose yes, although the data rate is *terrible*
<sorear>
and the range isn't that great
<kc8apf>
next year's flashing light content: blink a light at 1Hz using a nuclear power source
<sorear>
there's a proposal to site the second Hyper-Kamiokande tank in South Korea instead of the current Japanese facility and I'm certain geopolitics is playing absolutely no role in the process
onchip has quit [Quit: Page closed]
daveshah has quit [Read error: Connection reset by peer]
daveshah has joined ##openfpga
<mithro>
Has anyone seen / got a tool which converts between verilog specify files and sdf format?
unixb0y has quit [Ping timeout: 260 seconds]
unixb0y has joined ##openfpga
<rqou>
azonenberg: ping?
Bike has quit [Quit: Lost terminal]
<awygle>
I have met some traders working on free space optical links
<rqou>
just to be a little bit faster than fiber?
ZombieChicken has quit [Quit: Have a nice day]
<awygle>
Yup. Because 0.999c > 0.67c
<awygle>
(approx.)
<awygle>
also "blink a light at 1hz using a nuclear source" sounds pretty doable
<awygle>
tbh
<rqou>
so azonenberg didn't want to do open-source nuke-to-blow-people-up, but maybe we can do open-source nuke-to-make-energy? :P :P :P
<awygle>
I mean the crappiest rtg could blink an led at 1hz
<awygle>
(probably)
<rqou>
i mean, i bet even the nurdrage nuclear battery hack can blink an led
<awygle>
Tritium was my first thought
<awygle>
Americium was my second
<awygle>
In both cases I immediately got carried away designing direct conversion cells instead of PV or TC
* awygle
loves nuclear energy
<rqou>
but but but don't you know, terrorists could use them to cause massive damage!oneone :P :P :P :P :P
<rqou>
amidoinitright?
<awygle>
Only if they attack the weak point
<rqou>
i have no idea what current anti-nuclear FUD is like
<awygle>
yeah idk either
<rqou>
a bunch of people on birbsite were talking about it, so some group must have said something
<awygle>
Greenpeace EU yeah
<awygle>
Something about flying a plane into a reactor being way easy to do
Zorix has quit [Ping timeout: 276 seconds]
Hamilton has joined ##openfpga
<azonenberg>
greenpeace: environmentalism's worst enemy since $YEAR
* azonenberg
wonders how many tons of CO2 emissions could have been avoided if they hadn't been campaigning so hard against nuke plants
Zorix has joined ##openfpga
<rqou>
azonenberg: so, can i get a mini-design review?
<awygle>
looks design rule driven. are those all the signals you need?
<rqou>
yes
<rqou>
also, i'm being a bit conservative and designing for 6/6 even though the oshpark process is 5/5
<rqou>
and vias are definitely larger than minimal, but when i did do min-sized vias they always looked a bit sketchy on oshpark's process
<awygle>
it doesn't look terrible offhand, couple things I'd do differently but just stylistic
<awygle>
you're not going to take my advice and use seeed or dirty or jlc?
<rqou>
probably not
<rqou>
honestly i trust those fabs even less
<awygle>
fair enough
<awygle>
I've never actually had a problem
<awygle>
but I understand the concern
<awygle>
the cost difference is like 5x though hence the recommendation
<awygle>
(not even counting the greater quantities)
<rqou>
hmm i'll check
Hamilton has quit [Quit: Leaving]
<rqou>
given that this isn't _that_ demanding a board
<awygle>
shouldn't be no
<rqou>
i think i'll probably stick with conservative design rules iether way
<awygle>
sure, good not to push the envelope if you don't need to
<awygle>
i'ma sleep now, have fun routing
<rqou>
from my surveying oshpark seems to have the smallest min hole size of all these cheap fabs
<awygle>
that's very true
<rqou>
but it seems oshpark is pushing the fab a bit with those dimensions
<rqou>
i've noticed via drills looking like they're just barely not breaking out
<azonenberg>
rqou: i normally use 550/300 um trace/space as my standard via
<azonenberg>
sorry i meant
<azonenberg>
550/300 pad/drill
<rqou>
wtf i don't know um
<azonenberg>
then drop to 500/300 or 460/300 if i really need it in a tight spot (thats their design rule limit)
<rqou>
i'm using 12 mil drill 24 mil pad
<azonenberg>
they do 10 mil drils
<rqou>
i know
<rqou>
it always looks sketch af
<azonenberg>
I use oversized pads to reduce the risk of breakout
<azonenberg>
457 um is their design rule limit
<azonenberg>
i use 500 in tight spaces and 550 everywhere else
<rqou>
or maybe that's because minimum annular ring is a bit too small
<azonenberg>
yes, the annular ring spec is really tight
<azonenberg>
the drills are fine
<rqou>
also, many other fabs don't do down to 10 mil
<azonenberg>
but i think they're exaggerating the annular ring capability
<azonenberg>
its way smaller than any other cheap fab
<rqou>
that's what i've noticed
<azonenberg>
so i only use their full-spec annular rings when i have no choice in really tight spots
<rqou>
i mean, most of my boards work but always make me nervous
<azonenberg>
and i do it in a handful of spots on the pcb where it wouldnt be the end of the world if one failed and i had to rework
<rqou>
i don't get why they advertise such tight annular rights
<rqou>
*rings
<rqou>
hobbyists don't need it, and people like you who do will know they're cheating
<azonenberg>
its possible the fab is exaggerating their specs to oshpark? :p
<azonenberg>
idk
<rqou>
huh
<rqou>
i thought laen would be much more strict than that
<azonenberg>
i meant, the fab is lying
<azonenberg>
not laen
<azonenberg>
I have had no problems pushing limits on trace/space though
<rqou>
i thought laen likes to torture-test fabs
<azonenberg>
He does
<azonenberg>
but defects still slip through
<azonenberg>
i've had one board with totally borked soldermask due to a gerber merging bug
<azonenberg>
i've had one with 450um drills where i called for 300 because the fab screwed up and loaded the turret wrong in the drill machine
<azonenberg>
i had one with a very weird ENIG defect
<azonenberg>
And at least one with dust specks or similar on the photomask (the one you caught)
<rqou>
i know we went over this, but i'm still amazed just how many fab issues you seem to hit
<rqou>
azonenberg: i'm curious what your working relationship with laen is like? does he love you or hate you that you always encounter weird problems?
<azonenberg>
i make a lot of boards :p
<azonenberg>
and we havent talked in a while actually
<azonenberg>
since he stopped hanging out in #oshpark
<azonenberg>
i've had more contact with pdp7
<rqou>
i wonder what the average "complexity level" is of boards people make at oshpark
<rqou>
apparently there's a nonzero amount of non-circuit boards (art, mechanical, etc)
<azonenberg>
yeah
<azonenberg>
i've actually considered using pcb for front panels
<azonenberg>
you get nice silkscreened labels
<azonenberg>
optional conductive elements for ground connections or indicator LEDs
<azonenberg>
And milled cutouts
<rqou>
yeah, definitely not a bad idea
pie__ has joined ##openfpga
<rqou>
azonenberg: TIL BGAs are actually pretty nice to route as long as you don't actually need all of the pins
<rqou>
doing an escape is pretty relaxing and is a nice puzzle
<azonenberg>
yeah
<azonenberg>
and if you have enough layers, using all the pins isnt a big deal
<azonenberg>
(you see why i prefer them to trying to fan out a QFP?)
<rqou>
i have one part here that's a qfp with exposed pad and it's just frustrating AF
<rqou>
the other qfps are ok because i either optimized their pinouts or they're only using like a quarter of the pins
<daveshah>
mithro: it is primarily used to parse the sdfs that come from icecube to build the timing database
<mithro>
Oh it does have a reader at line 208
<daveshah>
I did play about the other day using it to get some cell timings for the ecp5
<daveshah>
But it seems Diamond specifies some stuff slightly differently
<daveshah>
ISTR it combined setup and hold together unlike icecube
<mithro>
daveshah: I would really like to get yosys to take verilog specify and sdf files and convert them to attributes on objects
<daveshah>
mithro: yeah
<mithro>
daveshah: I would really like to get yosys to be able to do equivalence check when timing is involved in the future too :-P
<daveshah>
mithro: that's a totally different problem domain to what Yosys does
<daveshah>
I'm not even sure what you would check against?
<mithro>
daveshah: I'm not sure I know what *I* mean - but for example a AND gate with unequal delays on inputs can dramatically affect the actual output -- were as when they have equal delay on the inputs you could specify it as an input or an output and the results would be exactly the same
<daveshah>
mithro: this feels quite far from anything Yosys is, and a very niche problem
<daveshah>
You would be in the realms of symbolic algebra I think, in the general case
<mithro>
daveshah: I guess we see Yosys as kind of different things -- I see yosys as a tool which takes descriptions in one format and converts it to another format while providing checks that the conversion and descriptions are valid
<mithro>
daveshah: Take converting adders to netlists -- there are many different ways to implement an adder with have different trade offs
<daveshah>
mithro: yes, exactly
<daveshah>
I understand what Yosys is
<mithro>
daveshah: In theory you should be able to use logic equivalence to /prove/ that they perform the equivalent
<daveshah>
mithro: yes, that's a common use case for Yosys
<daveshah>
But none of this ties into talking about equivalence checks on numerical timing values, which is quite different
<mithro>
daveshah: However it doesn't matter if they are logic equivalence if by using one form you can't meet your timing or logic usage requirements
<daveshah>
mithro: Remember that Yosys has no timing support for the fpga synthesis case, and very limited support in the ASIC case using ABC
<daveshah>
After all interconnect delays often dominate logic delays, so are unknowns at synthesis
<daveshah>
Ultimately, you use equivalence checks and formal verification to make sure it functions, and STA to make sure it meets timing
<daveshah>
I don't think anyone is interested in trying to combine the too, except for limited verification of clock domain crossings
<mithro>
daveshah: At the moment we generally force humans to make the decision, it would be nice if there was a much easier way for the systems to make the decision
<daveshah>
mithro: it's not exactly a difficult problem, it is a single logical and
<daveshah>
design OK = formal verification passes and STA passes
<mithro>
daveshah: Maybe Yosys will end up being the wrong place to add this, but eventually it will probably be wanted somewhere
<daveshah>
mithro: what do you actually want to add? What inputs and outputs do you expect
<mithro>
daveshah: The AND case was a trivial example, like printing the numbers 1 to 10 to show how a for loop might work
<daveshah>
What are the inputs and outputs you expect in the AND gate case?
<mithro>
daveshah: I want to know when two things are logically equivalent what the timing conditions I can do the substitution
<daveshah>
mithro: I still don't follow. I need an example of what you actually expect this tool to do
<daveshah>
mithro: yes, it seems quite far away from what Yosys does at the moment
<mithro>
daveshah: So yosys does the techmapping at the moment, right?
<daveshah>
mithro: yes
<daveshah>
But there is no timing at all involved in that
<daveshah>
And there has never really been a case where that is needed
<daveshah>
Given Yosys knows nothing about routing delays, its very hard to make any timing decisions
<daveshah>
They often dominate in FPGA synthesis anyway
<daveshah>
I'm not convinced that in the equivalence check there is any benefit compared to conventional equivalence check plus STA
<mithro>
daveshah: true -- but just because things are a certain way _today_ doesn't mean they should be that way forever more, I'm not expecting yosys to grow timing support today, or even probably any time soon - and I'm not even sure it is the right place
<daveshah>
mithro: this feels more like a PAR problem than a synthesis problem
<daveshah>
Yosys effectively has no idea what the delays at the A and B input would be
<mithro>
daveshah: You might be right
<mithro>
daveshah: First step is getting the data from the files into parameters however